hex4def6

The author explicitly shows how in Apple's walled garden, even that had hurdles that have been put up, 'for security'.

If you're benchmarking something, old & well-characterized / understood often beats new & un-characterized.

Sure, there may be shortcomings, but they're well understood. The closer you get to the cutting edge, the less characterization data you get to rely on. You need to be able to trust & understand your measurement tool for the results to be meaningful.

Aren't human coders also nondeterministic?

Assigning different agents to have different focuses has worked for me. Especially when you task a code reviewer agent with the goal of critically examining the code. The results will normally be much better than asking the coder agent who will assure you it's "fully tested and production ready"

As someone who worked on kindle power consumption many years ago: One of the (by far) biggest consumers of power is the WiFi connection. It has to wake up and respond to the AP in order to not get disconnected every x seconds.

Off the top of my head, I think 'on' average power consumption was ~700uA without wifi, and about 1.5mA+/- with Wifi. This is from over a decade ago, so my memory is fuzzy though...

Obviously, page changes used relatively large amounts of power. I don't recall the exact amounts, but it was 100s of mA for seconds.

There is also an "every x pages, do a full screen refresh (black to white)" to fix up the ghosting issue that the article writer saw.

Same.

I've been exploring some computer vision recognition stuff. Being able to reason through my ideas with an LLM, and make visualizations like t-SNE to show how far apart a coke can and a bag of cheetos are in feature-space has been mind blowing. ("How much of a difference does tint make for recognition? Implement a slider that can show that can regenerate the 512-D features array and replot the chart")

It's helping me get an intuitive understanding 10x faster than I could reading a textbook.

No.

I'd argue that no one would really care given it's GCC.

But if you worked for GiantSodaCo on their secret recipe under NDA, then create a new soda company 15 years later that tastes suspiciously similar to GiantSodaCo, you'd probably have legal issues. It would be hard to argue that you weren't using proprietary knowledge in that case.

Two socks and two sets of googly eyes are required for this exercise. Attach googly eyes to socks. Insert hands in socks.

The article is literally about how the efforts to keep humidity down have resulted in the growth of these extremophiles.

That's like saying 'why give people calculators, when you can pull out a slide rule'

The whole point is that you are enabling the LLM through tool use. The prompt might be "Download all the images on the wikipedia article for 'Ascetic', and print them on my dot matrix printer (the driver of which only accepts BMPs, so convert as needed)"

Your solution using file / curl is just one part of the potential higher level problem statement. Yes, someone could write those lines easily. And they could write the wrapper around them with only a little more difficulty. And they could add the 404 logic detection with a bit more...

Are you arguing LLMs should only be used on 'hard' problems, and 'easy' problems (such as downloading with curl) should be done by humans? Or are you arguing LLMs should not be used for anything?

Because I think most people would suggest humans tackle the 'hard' problems, and let the tools (LLMs) tackle the 'easy' ones.

If I can start an agent and be able to walk away for 8 hours, and be confident it's 'smart' enough to complete a task unattended, that's still useful.

At 3 tk/s, that's still 100-150 pages of a book, give or take.

Yes. And murder is illegal. And yet, Al Capone was in Alcatraz on tax evasion charges.

I'm failing to see the smoking gun here.

There are two ways you could interpret "counterfeit".

1. Fake IC (identifies as FTDI 232 IC), fake cable (FTDI logo on it)

2. Real IC, fake cable (eg, I buy the FTDI IC and make the cable, and sell it as an "official" FTDI cable).

(1) is I assume what they mean in this instance., but you could argue (2) is also possible. However, they make no mention of the packaging both calling them "FTDI" cables. Instead, I assume they're going off what they report to the OS as.

FTDI have been around for decades, and the offhand "old cable we had kicking around" could easily mean its 15+ years old. That might easily explain the chip size difference. In this case, FTDI did make TSSOP 28-pin chips for a long time. They're now obsolete, superseded by SSOP package variants (like in the "Real" picture). Put another way, this is like comparing an i5-10400 to a Pentium II that I found in my storage closet and declaring the Pentium II fake.

The actual fake chips visually look identical to the real ones. Obviously, otherwise they wouldn't get mixed into the supply chain.

The only real conclusion they can realistically make from these x-rays are that they're not the same cable (but even then, I don't know if FTDI real cables have silently upgraded the internals while retaining the same SKU).

If you're talking about Chavez, I'd disagree quite strongly. But even Maduro had his western-based supporters, at least in the early years; for example this op-ed in the Guardian:

  "Although there are abuses of power and problems with the rule of law in Venezuela – as there are throughout the hemisphere – it is far from the authoritarian state that most consumers of western media are led to believe. Opposition leaders currently aim to topple the democratically elected government – their stated goal – by portraying it as a repressive dictatorship that is cracking down on peaceful protest. This is a standard "regime change" strategy, which often includes violent demonstrations in order to provoke state violence."

https://www.theguardian.com/commentisfree/2014/mar/04/venezu...

We are discussing UI/Icon design, not the geopolitical implications of AGI or the Holocene extinction event.

Why should someone that disagrees with you on whether capitalism is uniquely responsible for bad icon design now be forced to defend it for every sin / shortcoming ranging from the social inequity to ecological collapse?

> We used to have a training ground for junior engineers, but now AI is increasingly automating away that work. Both studies I referenced above cited the same thing - AI is getting good at automating junior work while only augmenting senior work. So the evidence doesn’t show that AI is going to replace everyone; it’s just removing the apprenticeship ladder.

Was having a discussion the other day with someone, and we came to the same conclusion. You used to be able to make yourself useful by doing the easy / annoying tasks that had to be done, but more senior people didn't want to waste time dealing with. In exchange you got on-the-job experience, until you were able to handle more complex tasks and grow your skill set. AI means that those 'easy' tasks can be automated away, so there's less immediate value in hiring a new grad.

I feel the effects of this are going to take a while to be felt (5 years?); mid-level -> senior-level transitions will leave a hole behind that can't be filled internally. It's almost like the aftermath of a war killing off 18-30 year olds leaving a demographic hole, or the effect of covid on education for certain age ranges.

I think the forced countdown is a bit unfriendly, especially when LLM delays eat the majority of that. Perhaps limit it to X questions.

Also, having a min / max funding range (e.g, min 100K, max 500K). I might not want to wager the same on each opportunity.

It might also be interesting to base it off real companies. Scramble the names (Paypal -> CashMate or whatever. Enough that the player doesn't know for sure what company its based on. ATi vs nVidia for eg), and also allow investment at multiple points in time. Pre-crash vs post-crash might yield significantly different ROI outcomes.

You don't need a universal tool. You just need one that resets the drives you're selling.

I've worked with a vendor who were a bit fast and loose with what NAND / controller / firmware they considered "ACME SC9000" SSDs to be. Because of this, some of the drives actually had bad configurations. They gave us tools to query / reset / update the firmware on these drives. The SMART data was one of the options you could reset.

Given the number of $10 self-reporting "10TB" USB drives out there, if there's enough of a profit incentive and volume of drives, you can't rule out a SMART reset drive.

> Intel didn't. Solidigm did. If the author was buying the Intel drive, it was at least 4 years old since they spun SSDs out in 2021.

Firstly, the time between the SK Hynix acquisition (Dec 30 21) and the date of this article is 3 years 4 months, not "at least 4 years".

Secondly, of whether the facility was owned by Intel or Solidigm at the time the drive was manufactured, the Intel PCN states last buy dates of Dec 30 2022 here: https://www.intel.com/content/www/us/en/content-details/8055...

  The *Intel* SKUs listed in the products affected table will End of life ... Please determine your remaining demand for the products listed in the "Products Affected/Intel Ordering Codes" ... While *Intel* will make commercially reasonable efforts to support last time order quantities ...

It's entirely possible they did a large last-time factory build of drives in anticipation of people wanting to purchase them.

Or, as Solidigm state on their FAQ:(https://www.solidigm.com/support-page/faqs.html):

  Why do some Solidigm products have *Intel labels*, order forms, and branding?
  Certain business elements that were already in place or in development prior to the creation of Solidigm will continue to bear Intel labels and branding for some time. 

It's probably that the drives would have been branded "Intel" significantly beyond the Intel / Solidigm acquisition date (Probably until their EOL which was a year later -- it would make no sense to rebrand them). And it seems entirely unreasonable to assume that even a fairly tuned in customer would be digging to that level of scrutiny ("Wait a second! This is still Intel branded! Solidigm rebranded this line in XX of '22, X months before they discontinued them. These must be used drives!")

First of all, neither of those WILD facts seem that wild to me.

Intel did last orders for that drive Dec 30 2022. The article was written in April, so the author was conceivably purchasing drives that had sat on a shelf for a year and a half. That doesn't tickle alarm bells in my head.

Secondly, maybe my scam detector isn't well tuned enough, but "Maestro Technologies" doesn't seem that much stranger than "Apple" or "Micro soft" or "Zoom" or "Snap." If it were XBBHHZZZAA, LLC, maybe I'd have more room for pause.

The takeaway lesson here is that Amazon has become less and less reliable as a source for items. It's especially bad if it's purchased from a third party (something Amazon seems keen not to highlight on the purchase page), but even FBA is not free of trash. They straight up sell pirated N64 cartridges for example: https://www.amazon.com/Cartridge-Nintendo-Smash-64-Video-Ver...

This seems like a job for a truecrypt style system. Either you do it at a file-level, or you have it split into (say) 10MB file chunks, and if you want to access a certain file you have an encrypted local db that acts as a magic decoder ring ("file test.csv is spread across CLOUD1.DB CLOUD3443.DB CLOUD132.DB").

Combine that with steganography (Enter real_password, and test.csv is a list of bank accounts, enter fake_password, and test.csv is a list of apple store locations, enter random_password, and it decodes junk). Maybe combine that with multiple layers of passwords (one ring to rule them all, except certain files).

Obviously, you'd want to steganographize the decoder ring as well.

Because it's a strategic issue. The internet is critical infrastructure. While TP-Link might not have contracts with ISPs and datacenters, it doesn't take a lot of imagination to think what damage you could have with 30% of the home / small business routers under your control.

This could range from plausible deniability stuff (like the examples in the article), to targeted investigations / attacks (Bob who works at the Gov Accounting office for Miliary Spending), all the way to a 100-million unit botnet turning to provide a few days of distraction ("Bad hackers compromised our OTA system. Sorry!") on while a certain island is being eminant-domained.

Your food example is not the same. You can't trojan-horse an apple pie, or target an individual customer from the supplier-side (yet). If you decided to poison them, that's pulling the pin from the grenade right now.

For the vast majority, they're not "open source" they're "open weights". They don't release the training data or training code / configs.

It's kind of like releasing a 3d scene rendered to a JPG vs actually providing someone with the assets.

You can still use it, and it's possible to fine-tune it, but it's not really the same. There's tremendous soft power in deciding LLM alignment and material emphasis. As these things become more incorporated into education, for instance, the ability to frame "we don't talk about ba sing se" issues are going to be tremendously powerful.

It seems like the sensible thing to do would be to fry / erase any IFF and encryption related stuff, but otherwise continue as before.

E.g, if it's already been programmed to fly straight and level, continue to do that. If it's deactivated, stay deactivated.

Just seems like a whole 'nother set of characteristics to test otherwise, as well as adding extra unpredictability. The aircraft is probably damaged / on fire, so its flight characteristics are already going to be extremely different to normal. The best thing in the moment may be to let the aircraft lawn-dart in a field, rather than attempt to get straight and level, and in the process potentially fly over inhabited area or towards a friendly set of aircraft / buildings / vehicles.

Yeah.

I mean, why pay the money? Why not just skip the payment and email a contact "1,000"? Or perhaps "Interesting article about in the Times about the USA, wink wink"?

This method is deliberately communicating information in a way that (I assume) is prohibited. It doesn't seem like it would take a judge much time to come to the conclusion that the gag order prohibits communication.

Creating a secret code is still communication, whether that's converting letters A=1, B=2, sending a video of someone communicating it in sign language, a painting of the country, writing an ethereum contract, everyday sending a voicemail with a list of all the countries in the world from A to Z, but omitting the one(s) that have the gag / warrant...

Many years ago, I managed to stab my face with a screwdriver (not my proudest moment), and had to go to the ER. After the stitches, I was asked whether I wanted to pay with insurance. If I did, it was something like $2,000. If I didn't, there was a 75% discount off MSRP. My deductible was like 25%, so it ended up basically being the same out of pocket either way.

The fact that there seems to be a 4x markup means makes me think insurance companies are in bed with these hospitals. If you can mark up prices arbitrarily high, the insurance "discount" is fake.

You had better luck than me. The San Jose PD only begrudgingly gave me a police report weeks after reporting it (needed it for insurance purposes), and told me a could get a copy of it a month later. I'd have to go to the records dept in person between the hours of 10AM - 2PM (email a copy? Are you crazy?).

So I did that, showed up. No other people there. Person behind the counter told me they were too busy, and I'd have to show up some other (unspecified) day.

So yeah, I'd like to trade PDs with ya.

Tangentially involved in that project.

One of the big issues with that phone was that in order to do dynamic perspective, you're having to run a 3D render at 60fps constantly. That's a huge power hog, and prevents you from doing many of the power savings techniques you otherwise could on a normal phone -- shutting down the GPU, reduced refresh rate, heck, even RAM backed displays.

For ungrounded / 2-prong outlet devices, yeah.

It's often noticeable if you have a point contact of metal against your skin; sharp edge / screw / speaker grill, etc. Once you have decent coupling between your body and the laptop, you won't feel the tingle / zap.

They're called Y-caps if you want to delve deeper into them and their use in power supplies.

I'm assuming the thought process behind the pedal->generator->traction motor is that you can keep the cadence in the ideal / most efficient range. Problem with mid drives is that (at least for me), I prefer lower cadence which means driving the hub drive at a less efficient range.

I don't hate it, but I'm curious to see what they think their target audience is.

Is it someone that wants to do a "last mile" from their car/train to office? Is it a car replacement? Is it a weekend farmers market family bike?

I feel like all of these require emphasis on different aspects. If it's a last mile commuter, it needs to be light / small so it can fit on a train or car trunk. Probably also needs to be light so it's luggable into/out of the trunk by an average person.

I guess for me, the most interesting ebike idea is that of a "last mile(s)". Something that I can easily pack in the trunk, and depending on my mood, allow me to park ~5-7 miles from work and cycle the rest of the way in. There are bits of my commute near my office that are often at a standstill, and ebiking would be significantly faster to skip those. This is much less of a commitment compared to doing the entire 20mi commute by bike.

Yeah, agree. My concern would be a chicken sticking it's head through the gap to take a peek as its closing, and getting stuck.

I think the OP may have mitigation (or at least the possibility to mitigate).

This looks like an open-loop system (eg, the MCU doesn't know where in the swing the motor is), which makes it a bit more difficult. But it looks like they have limit switches.

Not quite sure how it determines the point to go to 100% power, but I assume it's a timing thing. I can't think of a good way to determine the difference between "chicken neck stuck in door" vs "snow / ice preventing door from closing" without some sort of position feedback.

I suppose you could have a timeout -- it gets 3 seconds at high power, and if it hasn't triggered the door close limit switch, it opens completely, then tries again. This would probably be ok, as long as 100% power doesn't decapitate the chicken...